Combination of hyperspectral imaging and machine learning models for fast characterization and classification of municipal solid waste

Abstract

Determining thermochemical properties and eliminating inorganic components of municipal solid waste (MSW) are crucial to its thermochemical treatment. Traditional characterization and classification technologies have shortcomings including long duration, complex operation, and inevitable sample consumption. This study proposed a hyperspectral imaging and machine learning models based method to solve these problems. Under the optimal parameter conditions, the identification accuracy of inorganic components by F1 scoring reached nearly 100% in MSW, and the prediction accuracy of carbon, hydrogen, oxygen, nitrogen contents and low heating value (LHV)of organic components by mean relative error value reached 92.6%, 86.9%, 80.4%, 54.7% and 90.5%, respectively. The results validated the hypothesis that combination of hyperspectral imaging and machine learning models are promising to accomplish fast characterization and classification of components in MSW, where principal component analysis was capable to abstract crucial information from the spectral pattern, and artificial neural network presented satisfactory classification and regression performance.